We’re Improving Leaderboard Accuracy ✅

[mal]

Good news! We’re updating our algorithms to make leaderboards more credible, and taking steps to proactively prevent suspicious efforts from appearing on leaderboards – so you can trust that the results you see are accurate. We're withholding from leaderboards any activities that appear incorrectly labeled (e.g. a bike ride getting uploaded as a run) or have faulty GPS data.

With these and other ongoing changes, we can better ensure your efforts will get the ranking they deserve and you can trust that the CR, KOM or QOM (plus all the other times on the leaderboard) are the real deal. These changes will be for all activities moving forward – we can’t yet capture past activities. And of course, you can still report an activity if you think there’s one the algorithm missed. Learn more about that here.

This is a big undertaking – we saw around 26,000 1-mile running splits that would have been faster than the world record last week! – and the work will be ongoing. Here’s how we’re committing to cleaning up leaderboards right now:

The Details 📝

✔️ There’s a new threshold for flagging: Activities with too much erroneous data will be automatically flagged – which means all segment efforts from those activities will be withheld from leaderboards.
✔️ We’re doubling down on catching bike rides (or downhill ski runs, car rides, etc.) marked as runs with new run-specific parameters that will flag activities based on their distance and pace data.

[susuncorda]

Great news.

Segments are a main motivation for many many riders as you know.

(We) The community helps by marking or tagging, but facilitating cleanup with tools that detect easy-to-identify cases is something that users appreciate.

Now all that remains is to fix the inequality of fighting in the classifications if you are an athlete riding alone VS a riders group (amateur or even a professional race) and I have contributed an idea about that in a separate topic.

[Alex_Michel]

Except their tools are not working as intended. Case in point:
https://www.strava.com/segments/18212621

[RonHuizenga]

It's nice to see this.  There are a lot of leaderboard items that don't even appear feasible.  case in point, the July Cycling Challenge - the leader has over 11,000 km in 25 days.  That's 450 km/day, every single day.  The Tour de France athletes are about 170 km/day an their active days.  There's a lot of work to analyze data, but it should be reasonably straight forward to add some validation by activity type for maximum feasible limits per time period to trap and exclude the outliers.  -  or is that out liars? 😉  

[Jan_Mantau]

These two "runs" were very fast but not in the impossible area as they were slower than world records. That can't be catched with algorithms.

[Alex_Michel]

What about that long stretch of running downhill at 1 min./Km? Also, anything "near" a world record should prompt the athlete, upon upload, to question their default activity type. Not put the onus on the larger community, which seems apathetic in my area, to flagging obvious abuse.

[zecanard]

You don’t even have to consider world records. Beating your personal record by 6 minutes per mile, over a course 3× as long as your previous best performance, should be an automatic flag, or at least prompt the user to confirm the activity type.

Plus, there are things like accelerometer data that can help determine whether an activity is a run. Syncing a run from a fitness tracker that includes very fast GPS data with no cadence information (from wrist-based arm swing or phone bounce) should be suspicious. I understand that phones may require additional permission to share motion data that not every user allows; however, more can be done on Strava’s end to legitimize activities.

[Alex_Michel]

✔️We’re doubling down on catching bike rides (or downhill ski runs, car rides, etc.) marked as runs with new run-specific parameters that will flag activities based on their distance and pace data.

Pretty sure this is not working as intended.

Effort thresholds such as sub-30 min. 10k, sub-15 min. 5 k, 4 min. miles and sub 1 min/km downhill runs (obvious rides uploaded runs), should be fairly easy to auto-flag.

Activities below were both recorded and uploaded after this alleged improvement was put into place :

https://www.strava.com/activities/9455702660/overview

https://www.strava.com/activities/9464929005/overview

Above are only two of several activities that triggered multiple "Un oh!" notifications for myself and several other athletes, claiming top spots on several leaderboards.

[zecanard]

With sections hitting 1 minute per mile, and 40-second sections under 3 minutes per mile, it’s clear that the auto-flagging algorithm needs work. Extremely obvious cheats, especially when you look at Paul’s previous activity, which is a “Run” that was later corrected to a bike ride. Then you look at Corina’s previous 3-mile runs at ≈11:20/mi, and it’s clear that she used the wrong activity type. She, too, has a “Morning Run” that she later corrected to a bike ride.

[Alex_Michel]

But yet here we are, after Strava is "... doubling down on catching bike rides (or downhill ski runs, car rides, etc.) marked as runs with new run-specific parameters that will flag activities based on their distance and pace data."

[Ian]

I noticed something slightly different today - someone in my local area typically does daily "bike rides" with improbable speeds (some over 100 mph) causing some distortion of the Segment leaderboards.  None of these Activities seem to have been auto-flagged.  However, today I noticed that one of the individual Segments was marked as "excluded" due to suspicious high speed (it had a speed of 42 mph).  The "ride" still has multiple spurious KOMs which were not excluded so the revised algorithm still needs more work.  The user had a top speed of 52 mph up a 3% slope - such an infeasible speed should have automatically excluded any adjacent Segments (that is, without there being a pause in Activity that might have allowed a transition from vehicle to pedal bike).